Yuqing dfc8a1346b docs: add dev guidelines, rename refactor/ → design/
- docs/dev_guidelines.md: tooling (uv, ruff, pyright, pytest, structlog),
  project layout, code conventions, git conventions, testing strategy
- docs/refactor/ → docs/design/ — these are design docs, not refactoring notes
- Updated all cross-references in dev/issues/README.md and dev/todo.md
2026-03-19 19:55:11 +08:00

Agl-lite: A Minimal Workable Version of Agent Lightning

This project agl-lite which is a minimal workable version of the popular agentic rl project Agent Lightning (https://github.com/microsoft/agent-lightning).

here are some key changes:

  1. remove the dependency of litellm and build self-owned request gateway instead;
  2. remove the dependency of OpenTelemetry and the whole stack built upon it, such as the tracers of agents; instead, use gateway to collect request-response data during transfer and record them into data store;
  3. following (2), the organization of data is not based on span in opentelemetry, instead, the basic trajectory format is sequence of requests (with response);
  4. use kubenetes as the default agent runner (minikube for single machine), and move the retry control from data store to k8s controller, and simplify the rollout states

There are some local environment setup and configuration needed for this project refactoring in .local/.

S
Description
用于快速构建 AI Agent 的训练工具。|GitHub 镜像 17.7k · 🍴 1.6k
https://github.com/microsoft/agent-lightning Readme MIT 56 MiB
Languages
Python 98.1%
Shell 1.9%